[cap-talk] Capability accounting

[Norman Hardy]

On Jun 24, 2006, at 11:43 PM, Nick Szabo wrote:

>
> Norm:
>> You seem to assume that it is the user who makes this decision,
>> or is even aware of the decision.
>> Within an enterprise most buy decisions are delegated to
>> low level purchasing agents.
>> In my computer I would have low level software
>> purchasing agents that would seldom bothered me.
>
> This doesn't get around the problem I have posed, namely that
> the agent must somehow learn about the user's preferences and
> then translate them into economic or security decisions.
> A human purchasing agent must learn quite a bit about the
> boss's desires and the organization's needs before he can act
> effectively.  The human will base quite a bit of this on
> experience and empathy rather than detailed instructions.
> There is nothing more worthless than an employee who requires
> detailed instructions before he can get anything useful done.

I agree with all your points. Agents for some situations are easy  
however.
I can instruct an agent in my browser to pay up to 1 cent for any
link that I explicitly click on but 10 cents for any NY Times page I  
select.
Other demands by web servers will require my explicit attention.
I do have to understand how to do this at least once and that
is a burden a little bit like figuring out how to subscribe to a  
magazine.
Deciding on a click price is a bit like considering the price of a  
magazine subscription.
There is a cost in deciding the worth and a cost in learning to  
express policy.
It is certainly not a psychological cost on each click any more than
when I turn on the lights in my kitchen.


>
> But detailed instruction is just what a software agent requires.
> The software can't be expected to have such experience or
> empathy, and will often not even be able to acquire detailed
> instructions as they are often tacit preferences that the user can't
> readily articulate more easily making the decision(s) herself.

See the scheme at <http://cap-lore.com/Economics/meter.html>
for a scenario where we battled the real issues you raise.

> And even with the human agent large decisions must be reviewed
> by higher-ups and aggregates of those decisions audited.

Yes, the same in computers!
One danger is of 10,000 undifferentiated anonymous deamons
sucking CPU and RAM in my computer.
Most of them are not supporting anything I value.
Many are outright contrary to my well being.
This is going off on another difficult tangent however.

> It's not that software agents are always impossible, it's just
> that they cannot be waved off as a mere implementation detail --
> they are the heart of any resource allocation or security system
> based on actual user preferences, such as proposed schemes
> to mimic real-world prices: useful prices are based on actual
> preferences.

I agree at least in the case of resource allocation,
I have not seen useful designs for agents that make security decisions.
That may be possible.
I am not impressed with Microsoft's security zone policy but that
is probably not what you have in mind.

> Nick Szabo
> http://szabo.best.vwh.net/
> http://unenumerated.blogspot.com/
> _______________________________________________
> cap-talk mailing list
> cap-talk at mail.eros-os.org
> http://www.eros-os.org/mailman/listinfo/cap-talk